Exercise Set 10 c 2006 Felleisen, Proulx, et. al.
10 Sorting, Performance/Stress Tests
In this problem set you will examine the properties of the different algo- rithms we have seen as well as see and design new ones. The goal is to learn to understand the tradeoffs between different ways of writing the same program, and to learn some techniques that can be used to explore the algorithm behavior.
Etudes: Sorting Algorithms To practice working with the ArrayList start by implementing the selection sort. Because we will work with several different sorting algorithms, yet want all of them to look the same the selection sort will be a method in the class that extends the abstract class ASortAlgo. A sorting algorithms needs to be given the data to sort, the Comparator to use to determine the ordering, and needs to provide an access to the data it produces. Additionally, the ASort class also includes a field that represents the name of the implement- ing sorting algorithm. abstract class ASortAlgo
/∗∗ ∗ Initialize a data set to be sorted ∗ with the data generated by the traversal. ∗ @param the given traversal ∗/ abstract public void initData(Traversal
/∗∗ ∗ Sort the data set with respect to the given Comparator ∗ and produce a traversal for the sorted data. ∗ @return the traversal for the sorted data ∗/ abstract public Traversal
1 c 2006 Felleisen, Proulx, et. al. Exercise Set 10
1. Design the class ArrSortSelection that implements a selection sort and extends the class ASortAlgo using an ArrayList as the dataset to sort. The ArrayList becomes an additional field in this class.
• The initData initializes the ArrayList by adding to it all the data generated by the given Traversal. • the new method selectionSort implements a mutating selection sort on this ArrayList. There are numerous lecture notes and labs from the past several years available online that guide you through the design of selection sort. The third part of you draft textbook has a section on selection sort on pages 102-112. • The sort method invokes a selection sort on this ArrayList and returns an instance of ALT the Traversal over the data in an Ar- rayList. • Do this once you include the code in the main part of your assignment. Include in the class a self test in the form of a method testSort() that provides a test for all methods in this class. There is an ex- ample of this technique in the AListSortInsertion class provided with this homework.
2. Design the class ArrSortInsertion that implements a mutating inser- tion sort for an ArrayList and extends the class ASortAlgo in that same way as was done for the selection sort.
You may work on this part with your partner if you wish — or get any help you need. However, you must make sure that at the end you understand every part of the code and how the tests were designed.
Sorting out Sorting We have seen by now three different ways to sort a collection of data: The insertion sort, the quicksort, and the binary tree sort. The binary tree sort was hidden — we first designed the method to in- sert data into a binary search tree, then we designed the traversal over the binary search tree that generated the elements in the desired order. The algorithms all have the same purpose, yet they do not conform to the same interface. They also deal with the data structured in different ways. However, we would like to compare these algorithms for efficiency
2 Exercise Set 10 c 2006 Felleisen, Proulx, et. al.
(speed).
ASortAlgo class: Reporting the results Our first task is to design wrappers for all these algorithms that will al- lows us to use them interchangeably to sort any collection of data supplied through an iterator. Of course, we want all of them to produce the data in a uniform format as well. Therefore, we want all of these algorithms to produce a traversal for the sorted list. Because the traversal of a binary search tree involes quite a lot of com- putation, to make to comparisons fari, we will conclude the binary tree sort by producing an sorted ArrayList, generated by the data produced by the traversal of the binary search tree we have constructed.
ASortAlgo class: Initializing the data The abstract class ASortAlgo introduced in the Etudes provides a uni- form wrapper for all sorting algorithms. The initData method consumes the given traversal for the data to be sorted and saves the given data in a data structure appropriate for this algorithm. For the algorithms that are based on the ArrayList the initData method creates a new ArrayList and adds all elements generated by the given Traversal. For the AListSortInsertion class that implements the insertion sort for a recursively built AList, the initData method copied the given data into a new AList. For the quicksort algorithm that is based on recuesively build lists, this will not be necessary - as the algorithm traverses over the data and constructs two new lists - the upper and the lower partition, and does not use any properties of the lists. Finally, for the binary tree sort the task of constructing the binary tree comprises a substantial amount of work, and so should be done as a part of the sort method. Therefore, the initData method just copies the reference to the Traversal. We provide an example of a class that implements the insertion sort al- gorithm for data saved as AList
ASortAlgo class: Running the timing tests To measure the time each algorithm takes to complete its task, we will invoke each algorithm in three parts. First, we provide the data to sort and invoke the initData method. Next we start the timer, invoke the sort method and stop the timer when the method returns the result. Finally, we check that the data was indeed sorted and record the result. We will test each of the several sorting algorithms on several datasets
3 c 2006 Felleisen, Proulx, et. al. Exercise Set 10
(varying the dataset size and whether the data is in the original order as received from the database, or is randomized). For each dataset we will use several different Comparators — by name and state, by latitude, and by zip code. The results will be recorded in a uniform way, so that we may then look for patterns of behavior. Here is a summary of the algorithms you will implement. Please, use the names given below:
• ArrSortSelection — done as a Part 1 of the Etudes
• ArrSortInsertion — done as Part 2 of the Etudes
• AListSortInsertion — provided
• ABinaryTreeSort — modify the solution to homework 5
• AListSortQuickSort — to be done (modify 8.1 part 9 of homework 8)
• ArrSortQuickSort — to be done
10.1 Problem Design the method in the Tests class that determines whether the data gen- erated by the given Traversal iterator is sorted, with regard to the given Comparator. Later, you will need the same method in the class TimerTests.
10.2 Problem Complete the design the classes AListSortSelection and AListSortInsertion. Include in each class a self test in the form of a method testSort(Tests tests) that provides a test for all methods in this class.
10.3 Problem Design the class ABinaryTreeSort that that extends the ASortAlgo class. It performs the binary tree sort on the data supplied by the Traversal iterator. The sort method first builds the binary search tree from the data pro- vided by the iterator, then saves the data generated by the inorder traversal in an ArrayList or in an AListOfCities data structure.
4 Exercise Set 10 c 2006 Felleisen, Proulx, et. al.
Include in each class a self test in the form of a method testSort(Tests tests) that provides a test for all methods in this class.
10.4 Problem
Design the class AListSortQuickSort that performs the recursively defined quicksort on the data supplied by the Traversal iterator and producing an AListOfCities data structure. You will need a helper method to append two lists together. HtDP has a good explanation of quicksort. Include in each class a self test in the form of a method testSort(Tests tests) that provides a test for all methods in this class.
10.5 Problem
Design the class ArrSortQuickSort that that extends the ASortAlgo class. It performs the quicksort sort on an ArrayList. The ArrayList is initialized from the data supplied by the Traversal iterator. You may use any textbook or the web to find an implementation of this algorithm, but you are responsible for the correctness of your implementa- tion. Include in each class a self test in the form of a method testSort(Tests tests) that provides a test for all methods in this class.
Note: If you are having problems with this algorithm, go on to the Time Trials and finish this only if you have time left.
Part 2: Time Trials
All of the tests we designed as the part of our code sorted only very small collections of data. It is important to make sure that the programs work well for large amounts of data as well. It is possible to estimate the amount of time an algorithm should take in comparison to others. However, we would like to verify these results on real data, and learn in the process what other issues we need to take into consideration (for example, the space the algorithm uses, and whether the data is already sorted or nearly sorted).
5 c 2006 Felleisen, Proulx, et. al. Exercise Set 10
Test Data
The class DataSet represents one set of data to be sorted. It knows the size of the data set, whether it is a sequential subset of the original data or a randomly selected set of data. It provides an iterator that generates for the sorting algorithm all elements in this data set. The class TestData generates all DataSets we will use, so that we do not have to repeat this process, and also to make sure that all algorithms will use sort the same data. This way we can conduct ’controlled’ experiments — comparing outcomes when solving the same problem.
Timing Tests
The class TimerTests provides a framework for conducting timing experi- ments. It contains a field that is an instance of TestData so we do not have to read the file citiesdb.txt of 29470 items for every test. The method runOneTest runs one test of a sorting algorithm. It con- sumes a sorting algorithm (an instance of ASortAlgo) and an instance of DataSet. These two pieces of data determine what is the data to be sorted, how large it is, whether it is random or sequential, which algorithm is used, and which comparator is used. It runs the sorting algorithm with a stop- watch and produces the timing result. Finally, the method allTests runs timing tests for all the selected combi- nations of sorting algorithms, comparators, and datasets and collects the results into an ArrayList of Results.
10.6 Problem
Design the classes that implement the Java Comparator interface and allow us to compare two cities by their zip codes (class ComparatorByZip) and by longitude (class ComparatorByLongitude).
10.7 Problem
Study the design of the class Result that holds the results of the timing tests. For each test we want to remember that the name of the test (for exam- ple ”Insertion sort with ArrayList”), the size of the data that we sorted, whether it was sequentially or randomly selected data, and the time it took to run the algorithm.
6 Exercise Set 10 c 2006 Felleisen, Proulx, et. al.
The method runATest in the class TimerTests modifies the method runOneTest to produce an instance of the Result. Modify the method toString in the class Result to produce a String that represents the result formatted the way you would like to see the results.
10.8 Problem The method runAllTests that consumes an ArrayList of instances of SortAlgo- rithm, an ArrayList of instances of Comparators, and the instance of TestData, and runs the timing tests for each algorithm, using each of the compara- tors, using both, sequential and random data. The results are produced as an ArrayList of Results. Run all possible tests. Make a note of those that fail. Choose a few at a time (for example all algorithms and comparators for a given dataset) and record the results into a text file or into a spreadsheet.
10.9 Problem Look at the results of the timing tests and see what you can learn about the different ways to sort data. If one of the algorithms takes too much time or space, you may elimi- nate it from further trials on larger datasets. However, try to understand why that may be hapenning. You may also modify the way the dataset is initialized. For example, you may want to see how your algorithm performs on sorted data. Present both the numerical results, and your comments about the mean- ing of the results in a professionally designed format — possibly with charts. We care both about the results and about the way you present them and explain what you learned from them.
7